Backprop with Approximate Activations for Memory-efficient Network Training

Open in new window